For decades, banks have looked at customers’ payslips, account statements, credit histories and collateral to decide if one can repay a loan. Once the money is disbursed, they often have to wait and hope repayments arrived on schedule.
With the increasing adoption of AI technologies such as machine learning, natural language processing and large language models, however, lenders are assessing borrowers beyond the point of approval.
They are using AI to continuously monitor customers, predict financial distress before it becomes visible and intervene long before a loan turns into a non-performing asset.
CBK data shows the banking sector’s stock of non-performing loans (NPLs) rose by Sh21 billion in the first quarter of 2026, reversing part of the improvement seen last year.
NPLs rose from Sh674.4 billion in December to Sh695.4 billion by the end of March, pushing the default ratio to 15.6 percent from 15.4. The increase came despite lower interest rates, highlighting the lingering financial pressure facing families and businesses.
“Increases in NPLs were noted in the personal and household, trade, agriculture and manufacturing sectors,” CBK Governor Kamau Thugge said.
Rather than relying solely on historical credit records, AI systems can analyse millions of data points to estimate the likelihood that a customer will struggle with repayments.
“One of the biggest challenges is how to use AI and data analytics to tackle the issue of non-performing loans,” Caritas Microfinance Bank CEO David Mukaru said recently.
Lenders are using machine learning models – programmes trained to recognise patterns in data and make predictions or decisions – to classify borrowers into different risk categories.
They are not solely looking at bank statements or payslips but incorporating alternative data such as mobile money transactions, utility bill payments and merchant activity to build more complete financial profiles.
For many small businesses and informal workers, these digital footprints can paint a clearer picture of financial behaviour than traditional credit records.
“Non-financial data and alternative data from financial statements are becoming better in evaluating creditworthiness and minimising risk,” Absa Bank Chief Operating and Digital Officer, Julius Kamau, said.
Banks are also beginning to use AI as an early warning system. By tracking changes in a customer’s financial behaviour – such as declining income, delayed bill payments or stress affecting the sector in which they operate – AI models can alert lenders before a borrower defaults.
This allows banks to engage customers earlier, restructuring repayment schedules or offering temporary payment relief.
Mr Kamau said customers increasingly expect banks to tailor products to their circumstances, making investment in data science and AI increasingly important.
“Customers are telling us they want hyper-personalisation of products, and that can only be driven by us knowing them as much as possible,” he said.
Despite the growing capabilities of AI, banks are not handing lending decisions over to machines. AI can update a customer’s risk profile as new information becomes available, but the final lending decision rests with human judgment, a Thomson Reuters report says.
That human oversight has become important as concerns grow over the potential for algorithmic bias. This happens because the data used to train the models may be flawed, incomplete or reflects human prejudices.
Consumer advocates say fully automated lending systems could unfairly disadvantage borrowers from certain neighbourhoods or those outside conventional financial profiles such as informal traders and freelance workers.
